Add non-uniform grid regression tests for linear advection (UpwindScheme)

[utkuyilmaz1903]

Summary

Adds a regression test suite for linear advection with UpwindScheme() on non-uniform AbstractVector spatial grids. Uniform-grid convection tests already exist; they do not exercise per-point stencil weights, boundary upwinding on irregular spacing, or the non-uniform discretization path in DiscreteSpace.

This PR adds coverage only — no changes to scheme or discretization code.

Motivation

Non-uniform grid support for advection is an existing code path, but it lacked dedicated tests comparable to MOL_1D_Linear_Diffusion_NonUniform.jl. These tests lock in expected behavior before further work on non-uniform upwind (convergence, performance, additional BC types).

What's added

• test/pde_systems/MOL_1D_Linear_Convection_NonUniform.jl — MMS-based tests for u_t + v u_x = 0 with UpwindScheme()
• test/runtests.jl — new CI group Convection_NU

Test coverage

Area	What is checked
Extreme stretching	Symmetric, one-sided, and Chebyshev grids (up to ~1:1000 spacing ratio). Discretization and time integration complete without BoundsError, NaN, or blow-up.
Directional switching	Positive and negative constant velocity; spatially varying velocity; inflow Dirichlet + outflow Neumann BCs on clustered grids.
Accuracy & leakage	Translating sine MMS (weighted L2); uniform vs non-uniform error comparison (err_n < 5·err_u); short-time Gaussian mass conservation.
Interface integrity	AbstractVector grid routing (s.grid, s.dxs), no StepRangeLen fallback, Float32 grids, chebyspace, uniform vs vector dxs typing.

Implementation notes

• Tests use Dirichlet MMS boundary conditions from the analytical solution. Periodic BCs (u(t, x₀) ~ u(t, x_L)) are not used because non-uniform upwind currently rejects interface/periodic BCs in _upwind_difference. This is documented in-test by design, not a test oversight.
• Time stepping uses SSPRK33 with a CFL-limited dt based on minimum(diff(xgrid)).
• Default advection scheme is UpwindScheme(); approx_order = 2 for centered operators on the same grid.

Test plan

julia --project=test test/pde_systems/MOL_1D_Linear_Convection_NonUniform.jl
GROUP=Convection_NU julia --project=test -e 'include("test/runtests.jl")'

Related work

Part of the non-uniform upwind verification effort (proposal weeks 1–2): establish a CI-backed regression baseline for UpwindScheme on AbstractVector grids before convergence/performance suites and any scheme fixes.

[utkuyilmaz1903]

@ChrisRackauckas @xtalax Here is the regression test suite for the Upwind non-uniform audit.

[utkuyilmaz1903]

@xtalax @ChrisRackauckas I added a test block to check Periodic BC behavior on non-uniform grids with UpwindScheme

Passing u(t, x0) ~ u(t, xL) fails during the discretize step. The system routes the periodic condition as an interface boundary under the hood, which triggers this specific assertion at upwind_difference.jl:35
AssertionError: Interface boundary conditions are not yet supported for nonuniform dx dimensions

To document this current limitation while keeping the CI green, I wrapped the discretize call in @test_throws AssertionError and left the solve step commented out as a # TODO.

[xtalax]

Excellent looks good, my mistake. Yes, adding nonuniform support for interface bcs would be a good contribution, just make sure you understand their other use in joining disperate domains with different physics together. See the MOL_1D_Interface_Nonlinear.jl for this use case

[utkuyilmaz1903]

Thanks @xtalax I will look into it as soon as my final exams are over on Friday.

[utkuyilmaz1903]

@xtalax @ChrisRackauckas I've just rebased the branch. Could you please trigger the CI